You are on page 1of 14

# 1

## MASSACHUSETTS INSTITUTE OF TECHNOLOGY

Department of Electrical Engineering and Computer Science

## 6.341 Discrete-Time Signal Processing

Fall 2004

MIDTERM EXAM
Tuesday, November 9, 2004
′′
• This is a closed book exam, but two 8 12 × 11
′′ sheets of notes (both sides) are allowed.
• Calculators are not allowed.
• Make sure you have all 14 numbered pages of this exam.
• There are 4 problems on the exam.
• The problems are not in order of diﬃculty. We recommend that you read through all the
problems, then do the problems in whatever order suits you best. Note, too, that very
often the later parts of a problem can be done independently of earlier parts.
• A correct answer does not guarantee full credit, and a wrong answer does not guarantee
loss of credit. You should clearly but concisely indicate your reasoning and show all
relevant work. Your grade on each problem will be based on our assessment of your
level of understanding as reﬂected by what you have written in the space provided.
• Only the booklet supplied is to be handed in. No additional pages will be considered
in the grading.You may want to work things through on the scratch paper provided
ﬁrst and then neatly transfer the work you would like us to look at into the exam booklet.
Let us know if you need additional scratch paper.
• We will be following a grading strategy which focuses on your level of understanding of
the material associated with each problem. Speciﬁcally, when we grade each part of a
problem we will do our best to assess, from your work, your level of understanding.
Based on our judgement, each part will receive one of the following: E (EXCELLENT
UNDERSTANDING), G (GOOD), R (REASONABLE), M (MARGINAL) or U (UN­
SATISFACTORY). Where appropriate, we may also use grades of E/G, G/R, R/M, or
M/U so that there is a ﬁner scale.
On each part of each exam question we will also indicate the percentage of the total
exam grade represented by that part, and your numerical score on the exam will then be
calculated accordingly. (For example if a part represents 16% of the exam grade, then
E=16, E/G=14, G=12, G/R=10, R=8, R/M=6, M=4, M/U=2, U=0.)
This system (which we’ve used in the course many times in the past) is designed to focus
your attention and ours on your level of understanding of the material, rather than on
what calculations and miscalculations receive or lose points. Consequently, in looking at
If you honestly feel that we’ve misjudged it based on what you’ve written in the exam
booklet, then we’d be happy to discuss it with you.
2

## 6.341 Discrete-Time Signal Processing

Fall 2004

MIDTERM EXAM
Tuesday, November 9, 2004

NAME:

1 (a)
1 (b)
1 (c)
1 (d)
2 (a)
2 (b)
2 (c)
3 (a)
3 (b)
4 (a)
4 (b)
4 (c)
Total
4

Problem 1 (28%)
In ﬁgure 1-1 are two systems consisting of a compressor and an expander.

System A:

## System B: x[n] � ↑3 wB [n] � ↓2 yB [n] �

Figure 1-1:

(a) (4%) For x[n] as shown in ﬁgure 1-2 sketch yA [n] and yB [n] (assume x[n] = 0 outside the
interval shown).

x[n]
�b
a�
�c
� � � � � � �
0 n

Figure 1-2:
Name:
5

(b) (6%) For X(ejω ) as shown in ﬁgure 1-3, sketch Wa (ejω ) and YA (ejω ).

X(ejω )

1
��

� �
� �
� � �
−π − π2 π
2 π

Figure 1-3:
6

System A:

## Figure 1-1 repeated for your convenience.

(c) (8%) X(ejω ) denotes the Fourier transform for an arbitrary x[n]. Express YB (ejω ) in terms
of X(ejω ).Your answer should be in the form of an equation, not a sketch for a speciﬁc
Fourier transform.

(d) (10%) For any arbitrary x[n], will yA [n] = yB [n]? If your answer is yes, algebraically justify your
Name:
7

Problem 2 (28%)
Consider the system in Figure 2-1.

## x[n] � ↑4 w[n] � H(z) g[n] � ↓2 y[n]�

Figure 2-1:

We would like to implement the overall system (i.e. from x[n] to y[n] as eﬃciently as
possible in terms of the number of multiplies per output sample, the total number of compres-
sor/expander blocks, and the total number of delay elements.

1 − cz −1
(a) (8%) If H(z) = , where a and c are arbitrary and the system must be implemented
1 − az −1
exactly as shown in ﬁgure 2-1, with H(z) in direct form, determine the required number
of multiplies for each value of y[n].
8

## Figure 2-1 repeated for your convenience

For parts (b) and (c) assume that the cost of each multiply per output sample y[n] is 10,
the cost of each compressor or expander block is 5, and the cost of each delay is 2.
1 − cz −1
(b) (10%) If H(z) = , determine the ﬂowgraph or block diagram of the overall system
1 − az −1
which minimizes the implementation cost of the overall system in Figure 2-1, and the
total cost of the implementation.

Total Cost:
Name:
9

For parts (b) and (c) assume that the cost of each multiply per output sample y[n] is 10,
the cost of each compressor or expander block is 5, and the cost of each delay is 2.
1
(c) (10%) If H(z) = , determine the ﬂowgraph or block diagram of the overall
1−az −4 − bz −8
system which minimizes the implementation cost of the overall system in Figure 2-1, and
the total cost of the implementation.

Total Cost:
10

Problem 3 (20%)
A bandlimited signal xc (t) has been sampled and the samples have been stored in memory. The
samples are to be converted to a continuous-time signal yc (t) through an ideal C/D converter.
It is known that because of a hardware fault, the input to the D/C will be faulty. The objective
in this problem is to consider the possibility of pre-compensating x[n] in memory so that the
reconstructed signal yc (t) at the output will equal xc (t)

## xc (t)� x[n]� � � Compensated w[n] = � hardware g[n]

� D/C �H(jΩ) yc (t)�
C/D values in
� memoryx[n] +c[n] errors
� �

T c[n] T

Xc (jΩ) = 0, for |Ω| >
3T
w[n] = x[n] + c[n]

X(ejω ) = 0, for < |ω| < π
3

1 for |Ω| ≤ 3T
H(jΩ) =
0 otherwise

(a) (10%) For this part assume that c[n] = 0 (i.e. there’s no compensation and, consequently,
w[n] = x[n].) Assume further that the hardware errors have the eﬀect of adding an
undesired signal to w[n], i.e. g[n] = x[n] + e[n]. Under what constraints on e[n] or its
Fourier transform E(ejω ) will yc (t) = xc (t)? Clearly explain your answer.
Name: 11

(b) (10%) For this part assume that the eﬀect of the hardware error is to always force g[0] to be zero.
In designing the compensating signal we note that if c[0] = −x[0] then w[0] = 0. Then the
hardware error will not aﬀect w[n], i.e. g[n] = w[n]. With this in mind and considering
your answer in (a), determine an appropriate choice for the compensating signal c[n] so
that yc (t) = xc (t).
12

Problem 4 (24%)
For this problem you may ﬁnd useful the lecture slides which we have reproduced
on page 14.

## Consider the signal

� �n � �n
2 1
s[n] = α u[n] + β u[n]
3 4
where α and β are constants.

We wish to linearly predict s[n] from its past p values using the relationship
p

ŝ[n] = ak s[n − k]
k=1

## The coeﬃcients ak are chosen to minimize the prediction error

+∞

E= (s[n] − ŝ[n])2
n=−∞

(a) (4%) With φs [m] denoting the autocorrelation of s[n], write the equations for the case p = 2
the solution to which will result in a1 , a2 (you do not need to derive the equations).
Name: 13

(b) (10%) Determine a pair of values for α and β such that when p = 2, the solution to the normal

11
equations is a1 = 12 and a2 = − 61 . Is your answer unique? Explain.

(c) (10%) If α = 8 and β = −3, determine the reﬂection coeﬃcient k3 , resulting from using the
Levinson recursion to solve the normal equations for p = 3. Is that diﬀerent from k3 when
solving for p = 4? (An answer with no explanation will receive no credit.)

END OF EXAM

14

YOU CAN USE THE BLANK PART OF THIS PAGE AS SCRATCH PAPER BUT

L e v i n s o n  D u r b i n R e c u r s i o n

p a

[ ] = [ ] = 1 , 2 , K , p

k  s i  k  s i i

k = 1

[ ]

[ ] [ ] L

! s 0 ! s 1 ! s p ) 1

[ ] [ ] L

! s 1 ! s 0

T p =

[ ]

L [ ]

! s p ) 1 ! s 0

[ ] T r

[ ]

[ ]

[ ] [ ]

[ ] T I

[ ]

[ ] [ ]

[ ]

## p = a p , a p H 1 K a 1 p = < s p < s p J 1 K < s 1

( ) T ]

[ ] W X p p

k p + 1 = P s p + 1

( ) T i

[ ] c r p p

^ s 0

( ) ( )

a 1 p a p p

j p = m k p + 1 q p

( ) ( )

p p

a 2 a p ¡ 1

 p + 1 =  p  p

( 7 )
=

M £ k M

0 k p + 1

p + 1

( ) ( )

p p

a p a 1

£ 1

( ) a k t h

i s t h e c o e f f i c i e n t f o r t h e p t h o r d e r f i l t e r .

a k p